Generative AI is unlocking new ways to drive innovation, improve productivity and derive more value from data. For organizations to fully capitalize on this potential, it’s critical that everyone — not just those with AI expertise — is able to access and use generative AI. That’s why we created Snowflake Cortex (in private preview), Snowflake’s new, intelligent, fully managed service that enables organizations to quickly analyze data and build AI applications — all within Snowflake. As part of Snowflake Cortex, users of all skill sets now have access to industry-leading AI models, LLMs and vector search functionality, as well as complete LLM-powered experiences. These innovations enable all Snowflake users to securely tap into the power of generative AI and unlock dynamic insights with their enterprise data — regardless of their technical expertise.
Serverless functions in Snowflake Cortex
With Snowflake Cortex, Snowflake users now have access to a set of serverless functions that easily accelerate everyday analytics and AI app development. With just a single line of SQL or Python, analysts can instantly access specialized ML and LLM models tuned for specific tasks. They can also leverage more general purpose models for prompt engineering and in-context learning. Since these are fully hosted and managed by Snowflake Cortex, users always have access to them without the need to bring up and manage expensive GPU infrastructure. They can also use and leverage Snowflake’s unified governance framework to seamlessly secure and manage access to their data. These functions include the ones listed below.
Cost-effective LLM-based models that are great for working with unstructured data:
- Answer Extraction (in private preview): Extract information from your unstructured data.
- Sentiment Detection (in private preview): Detect sentiment of text across your table.
- Text Summarization (in private preview): Summarize long documents for faster consumption.
- Translation (in private preview): Translate text at scale.
- Forecasting (generally available soon): Train on historical time series data and forecast that time series into the future with automated handling of seasonality, scaling and more.
- Anomaly Detection (generally available soon): Identify outliers in your time series data for data pipeline monitoring and more.
- Contribution Explorer (in public preview): Quickly identify dimensions contributing to the change of a given metric across two different user-defined time intervals.
- Classification (in private preview soon): Categorize data into predefined classes or labels to better make recommendations based on patterns in the data.
State-of-the-art models that can be used for more general purpose use cases:
- Complete (in private preview): For given a prompt, the function returns a text completion response using cutting-edge, open-source LLMs such as Llama 2. See demo here.
- Text2SQL (in private preview soon): SQL is generated from natural language using the same Snowflake LLM that powers the Snowflake Copilot experience to help customers build their own applications.
This out-of-the-box functionality can be used for analysis, as well as part of app development in Snowflake. For example, these functions can be incorporated into a chatbot using Streamlit with just a few lines of code. This means anyone who knows Python can securely build powerful LLM apps in minutes or hours, not days or weeks.
Native LLM experiences built on Snowflake Cortex
Snowflake Cortex brings powerful AI and semantic search capabilities to the Snowflake platform. We have built several features to make the Snowflake user experience better by leveraging the power of Snowflake Cortex. These include pre-built user interfaces, high-performance LLMs, and search capabilities fully hosted and managed by Snowflake Cortex, making them ideal for business teams and analysts across organizations.
Snowflake Copilot (in private preview) is an LLM-powered assistant to generate and refine SQL with natural language. Analysts can ask Snowflake Copilot a question, and it will write a SQL query using relevant tables. Users can also refine queries through conversation to filter down to the insights most relevant to the task. No setup is required. In addition, this text-to-code functionality will be coming soon programmatically via a general purpose function, Text2SQL, with Snowflake Cortex.
Universal Search (in private preview) is an LLM-powered search for quickly discovering and accessing data and apps. Built on search engine technology acquired from Neeva, Universal Search helps you find database objects within your Snowflake account as well as data products and Snowflake Native Apps from the Snowflake Marketplace. With the initial release, you will be able to find tables, views, databases, schemas, Marketplace data products, and Snowflake documentation articles. Behind the scenes, Snowflake Copilot also takes advantage of Universal Search to identify relevant tables and columns for SQL generation.
Document AI (in private preview) is an LLM-powered experience for data extraction use cases. Using a pre-trained model and intuitive interface, customers can process any document (pdf, word, txt, screenshots) and get answers to their questions. This can be scaled up to become a pipeline to address extraction and save on resources in both manual labor and time. Announced at Summit in June, it is available now in Private Preview. See Document AI in action on YouTube.
Snowflake Cortex is putting industry-leading ML and LLM models into the hands of all Snowflake users via intuitive experiences and serverless functions, so they can quickly and securely get more value from their enterprise data. There’s no need for specialized AI expertise or a complex infrastructure to manage. Further, Snowflake Cortex provides the building blocks to create custom AI apps in the Data Cloud in minutes. To learn more about how you can build with Snowflake Cortex, read more in Fast, Easy and Secure LLM App Development With Snowflake Cortex. And to request access to private preview features, reach out to your account team at Snowflake.